DEV Community

power zhong
power zhong

Posted on

Tried `sponsors/calesthio` Today: An Open-Source Video Studio for AI Agents

Tried sponsors/calesthio Today: An Open-Source Video Studio for AI Agents

sponsors/calesthio is pitching something unusually ambitious: turn an AI coding assistant into a full agentic video-production studio.

The repository bundles 12 production pipelines, 100+ tools, and 700+ agent-skill and production-knowledge files. Instead of stitching together separate workflows for scriptwriting, storyboards, voice, assets, editing, and publishing, the goal is to let an agent coordinate the whole pipeline from a project brief.

It has gained +1,292 GitHub stars today, which makes sense. Video content is expensive not because any single step is impossible, but because every handoff adds context switching, SaaS subscriptions, and operational overhead. A local/open workflow that lets my existing coding agent drive repeatable production steps is a compelling indie-hacker play.

For a practical setup, I’d connect the agent’s Anthropic-compatible endpoint through a relay and make the production assistant use claude-fable-5 as its primary model:

export ANTHROPIC_BASE_URL="https://b-lost.com/v1"
export ANTHROPIC_API_KEY="your_b_lost_api_key"
export ANTHROPIC_MODEL="claude-fable-5"

# Example project instruction
agent run "
Create a 60-second product demo:
1. Write the script and shot list
2. Generate narration and visual prompts
3. Run the appropriate production pipeline
4. Export captions and a publishing checklist
"
Enter fullscreen mode Exit fullscreen mode

The architecture is straightforward: your coding agent handles planning and tool calls, Calesthio’s skills provide the production knowledge, and the model endpoint supplies the reasoning layer. That means you can keep the workflow close to your codebase, version scripts and assets, and automate videos alongside product releases.

For longer iterative runs, native Anthropic /v1/messages Prompt Caching is especially useful: production agents repeatedly carry large briefs, brand rules, scene definitions, and asset constraints. Caching stable context can reduce latency and avoid paying repeatedly for unchanged instructions. B-Lost’s Anthropic-compatible relay supports that native caching flow, while its 0.8x official pricing is a practical cost lever for bootstrapped teams.

This is still a system to test rather than a one-click replacement for a creative team. But for founders shipping demos, changelog videos, tutorials, and localized product content, it looks like a serious experiment worth cloning.

Top comments (0)